← Voices of Search // A Search Engine Optimization (SEO) & Content Marketing Podcast
Voices of Search // A Search Engine Optimization (SEO) & Content Marketing Podcast
G2 data: 1/2 of global software buyers now start search on AI chatbots instead of Google(71% jump since Apr); AI search is killing “Google first” era; Companies must rethink visibility in LLM-first era; Workplace adoption is consolidating around AI giants
00:53:00
Loading summary
Transcript68 lines
- [00:00]
Jordan Cooney
The Voices of Search Podcast is a proud member of the I Hear Everything Podcast Network. Looking to launch or scale your podcast, I Hear Everything delivers podcast production, growth and monetization solutions that transform your words into profit. Ready to give your brand a voice? Then visit iheareverything.com welcome to the Voices of Search Podcast. A member of the I Hear Everything Podcast network, ready to expedite your company's organic growth efforts. Sit back, relax, and get ready for your daily dose of search engine optimization wisdom. Here's today's host of the Voices of Search podcast, Jordan Cooney.
- [00:45]
79% of global B2B buyers say AI search has changed how they conduct research, proving that AI isn't a fringe tool anymore. It's front and center in how B2B buyers research software. That means 8 in 10 buyers are using AI to gather, compare, and shortlist vendors before they ever click a traditional search result. But most companies are still optimizing for the old playbook, climbing Google rankings and hoping prospects find them there. Here's the rub. Today's buyers are not starting with search. They're focused on AI chat. And that's the key result in success. So here's the real question. How do you win awareness in a world where buyers discover software through AI and not blue links? I'm Jordan Cooney, and joining me today is Tim Sanders, chief innovation officer at G2, home to one of the largest data sets on real B2B buyer behavior. G2 serves over 100 million software buyers annually and sees firsthand how enterprise research habits are shifting not gradually, but rapidly. Today, Tim will unpack what G2 data reveals about how buyers are using AI search, why traditional SEO no longer guarantees discovery, and what companies need to do to win in this new era of geo. Tim, welcome to the Voices of Search podcast.
- [02:01]
Tim Sanders
Glad to be with you, Jordan.
- [02:03]
Jordan Cooney
Awesome. Super stoked to have this conversation. Partly because you have a long history in search, right? I mean, we're in a new era, and this new AEO Geo, AI search era is certainly consuming most of the headlines. But you've got a lot of history with this. Why don't you tell us a little bit about how you got into search and some of your first experiences in the search world.
- [02:26]
Tim Sanders
Yeah, so I came by search kind of honestly. I was an early user, mid-1990s. Archie Gopher, Northern Lights, for those of you my age, that might ring a bell. And then the great disruptor Alta Vista. If you remember, um, I worked with Mark Cuban at his startup in Dallas, broadcast.com we were purchased by Yahoo in 1999. I went to Yahoo with the acquisition and by 2001 I became Yahoo CSO. And at the time, what I was really overseeing, Jordan, was the absolute disruption of the buyer journey as it went through things like direct mail, yellow pages, radio, TV and print. That's the first thing we saw just get absolutely disrupted by the birth of search mostly. Alta Vista became an inflection point for that first wave. And then we started to see digital advertising, which is my point of emphasis. Digital advertising, banners, promotions, that kind of thing. Also make a little bit of headway. Not a lot when I was there, but a little bit. But you could see the writing on the wall that the buyer journey was going to change and that effective ways of reaching them would change. Now I was there when Larry and Sergeant, you know, came to us with their amazing invention out of, you know, Stanford, which was PageRank and a real way to think about solving the keyword stuffing problem that currently existed. Making Search a very low quality experience for end users. I remember we had a chance to invest in them. By the way, don't believe the story that we could have bought them for a million. That is not the truth. The truth is we could have invested and then we should have invested in them. Why didn't we do it? No one had made any money at search. We thought it'd be better just to hire them as a vendor. But we saw what happened by like 2003, 2004, Google came in with a very simple, elegant, user focused solution and they stole share from everybody, including Yahoo. And then over the next 20 years, I have to say, Jordan, I watched the twists and turns of how PageRank evolved, how the rules could change very quickly. But what I tended to spend most of my time on as a consultant in my post Yahoo days was focusing on the things that don't change over time. Like what are the long standing practices you can have from a site structure and content standpoint that could create continuity in your search. Right, right. So it's like I do think a lot about what doesn't change so that we as marketers can make plans that are, that are durable. I did also see a lot of short term gamification that didn't work. Google was really built on the idea of solving for gamific, like keyword stuffing. So they've been very agile. However, the other thing I would say is much like everybody else, I became very dependent on Google Analytics and the ability to use databases to look up performance. And that's what's changed with answer engine optimization or generative engine optimization is. There is no great database yet to really look it up. You have to propagate it. And so that presents a whole host of measurement problems which are top of mind for me. Right?
- [05:43]
Jordan Cooney
Yeah, exactly. That's the beauty of this. You've seen this transition and you've seen these buyer journey transitions in the past. And I think that that's a really critical framing for many of our listeners, largely because the focal point here is a buyer. Right. It's a real person. And as you're seeing and as you're researching this, I'm curious to know more about how you see enterprise business decision making buyer journeys changing with the introduction of AI search and this whole Geo AEO revolution that's going on.
- [06:19]
Tim Sanders
Think about it this way. And I have to say I'm very focused on the B2B software buyer. So a lot of my research is pertinent to that. But I think directionally it will be consistent for you, even if you're a consumer and not software B2B. From the buyer standpoint, unless you're in procurement, which is a small percentage of buyers, unless you're in procurement, buying stuff isn't your job. Your job is like producing value and operational excellence and you have to buy stuff along the way, right? So it's like I buy office supplies, but office supply buying is not my job. So you have to understand that it's a tax on our time. And what that means is that tools that come along that are faster and make us more productive, we will adopt, and we will especially adopt those. If our scope of work is increasing. Maybe because we work at a company where they've had production enforced, they've had a headcount freeze, we're really getting squeezed to do more with less. So what's happened is the software buyer outside of procurement has moved from reference to inference. Let me unpack that for you.
- [07:23]
Jordan Cooney
Yeah, please.
- [07:24]
Tim Sanders
When you were buying software, it was like you went to the library of sorts. Now, the librarian would never do your job. The best they would do is point you to the stack of cards or point you to the periodicals or point you to certain reference areas for you to go look things up. You still had to do your job. That's Google. Google's the great librarian. And reference meant that you manually went to Google. You entered keywords, you got blue links, you begin to build a spreadsheet. You kind of created a short list eventually, and then the shortlist with you and the buying group, if there is one Helped you pick a winner and you bought that vendor. The process of just creating the short list from a standing start probably took you between 5 and 12 hours. That's what our research has seen at G2 over our buyer behavior reports. So it took somewhere between 5 and 12 hours. Kind of painful, and it required many days. If you looked at like the buyer journey from a calendar standpoint, a lot of that was that front end of all that reference work. Well, now they want to just do inference. What does that mean? Share the goal, share the pain. Let AI do all the synthesis and let it do inference. Now to give you the shortlist, and that's what's changed.
- [08:31]
Jordan Cooney
I got to ask you a question on that real quick because I love this and I do agree on this inference point, but like many of these models hallucinate. I mean, they, the founders, even Sam Altman, you know, stipulates that, hey, 33 to 44% of the time these models are going to hallucinate and provide something that's inaccurate. So, so it's, it's, it's changes.
- [08:52]
Tim Sanders
Let me tell you how this works, Jordan. It's like hallucinations are confident errors, which are often the confabulation of data. The number one way to reduce hallucinations is less training or structure of the models and more context that we give the models. So hallucinations have been as much a skills problem on the part of the human as it has been a model problem. Well, I'm going to walk you through some fun here. Okay. I'm going to go Sheldon Cooper Big Bang Theory on you now. So, you know, in addition to my work at G2, I'm an executive fellow at the Digital Data Design Institute at Harvard, and we geek out on stuff. So let me talk to you a little bit about hallucinations. So machine hallucinations have been coming down probably 90% a year since 2022, about as fast as the price of intelligence is coming down. So the number of hallucinations is dropping dramatically as we see new reasoning models, it drops even more because reasoning is iterative and quality prompts can reduce hallucinations by design. So someone who's got a good prompt skill can probably see continued 90% reductions hallucinations now, not every year, but every time a new model comes out. So it's reducing dramatically. When Altman said that, if said it again, he'd probably say 3 to 5%. It really depends on the use case, Jordan. It really does. Right? You know, so, so that's the issue. So, so that's the number one thing. The number two thing, though, however, is let's talk about human hallucinations. I remember talking to the president of a very large bank, and she said, you know, everybody came to her and said, like, like. But the machines hallucinate. She's like, have you met my people? Right. Humans make confident errors. Now, if you look at academic research around confident errors, like the ability for a person to believe something that's obviously not true, maybe subject to misinformation, maybe they got duped, maybe they're having a senior moment. Whatever it is, the human hallucination rate's been relatively stable from World War II when they started measuring a proxy for it, probably starting in 1955, I think is the first study that looked at, like, human errors that were persuasive and very much believed by the person, even though they were absolutely untrue and they weren't lies, there were just errors. It was pretty consistent, didn't really rise or fall. It kind of went up a little bit in the 80s and 90s with improvement in education, but just a little bit. And then after Covid, hallucinations have been on the rise. And then after misinformation from social media, now AI, they're actually rising, not falling. So there's inflection point coming, Jordan, and it ain't far away where the hallucination rate of a human is going to be statistically higher than a machine and not by a small margin. Interesting. So I think hallucination rates is a red herring that a lot of us get duped into as giving us comfort of saying, well, I'm not worried about AI because it's not accurate. My AI is very accurate because I use adversarial AI. So when I run, for example, when I run an audit on a website to see what its AI crawlability issues are, and I get a return, you should see the sophistication of my markdown prompt at this point based on a lot of iterations of it and talking to the sources. Then they say, no, that's wrong for this reason. And we tune and we tune and we tune. Get a really good markdown file for that prompt, and then I run it on ChatGPT5 Extended Thinking, and then I take the return, and then I go to Gemini 3 Pro and I say, this is your competitor's claim. Red flag everything that's not true. Recrawl it. And then we use that. My hallucination rate's almost zero on this stuff. So, like now when I go to a customer and say, here's our audit results. I feel about 90% of it's right. So, so we can.
- [12:28]
Jordan Cooney
We, we.
- [12:28]
Tim Sanders
I know it's a long answer for you, but don't take comfort in hallucination rates because that's, that was a. It's actually turning out to be more of a feature than a bug. It's. It's also the source of great creativity and scientific breakthroughs.
- [12:41]
Jordan Cooney
The same leads to. Let's bring this back to buyers and enterprise buyers specifically, or B2B buyers. You know which, which G2 helps here, right? Because if hallucinations is. Is. Is inherently a feature, which, which I definitely agree with you there. There's so much that we as humans need to understand to be great operators of AI as buyers and decision makers in this journey to make a software or a tool, a product decision that we want to make. How is it that G2 is leveraging AI or how are you thinking about leveraging AI to reduce that time? That main pain point that you highlighted earlier, which is the amount of time and energy that this person who typically isn't hired to do the job of buying, is now involved in this process.
- [13:31]
Tim Sanders
So let me give you an example. Using a mortgage loan. So think about like if you're a bank and you're approving a jumbo loan, which is not much different than buying a $200,000 a year CRM solution. It's a jumbo loan, high risk. There's probably 17 steps to approve a jumbo loan. Okay, Humans used to do all 17. And then with machine learning and RPA, now humans probably do like six of the 17. A lot of the other stuff gets organized, but it didn't quite maybe not even that many. Maybe they do 10 of the 17 still human. Why? Because RPA and machine learning is brittle. It doesn't apply to all the workflows. And then here comes large language models. They're very dynamic, they're stochastic, they coverage the whole thing. And so now what we're seeing in the world of mortgage is like at the end of the day, only step 17 is a human where they took all of the machine generated predictions and they pass judgment and that's, that's the way the world works. That's how software buying is starting to work. So the users in 2024, they started to use ChatGPT as a way of just avoiding the long Google session to start to get a granular list together so that they got that spreadsheet piece done in an hour or two instead of five. Hours, which was really the biggest bottleneck. And then starting in 2025, they just started on ChatGPT. So one of the things we learned, we did an April survey on this and we talked to B2B software buyers, good enterprise sample size, and 87% said that AI search, defined as a chatbot, not AIO reviews a chatbot. 87% says, changing the way I do research. And 29% said I start on the chatbot instead of Google now. We were like, wow, that's a big number. And then we did the survey again August 29, same, same end, same question, 50% so. So there's a move that started, but what we're seeing is that they no longer use keywords, they do best of prompts. I want you to give me the best three CRM solutions for a medium independent hospital that will have an endpoint on mobile devices because they have a particular use case name and they're one shotting the short list. So they're taking the hour and making it in minutes. And then that way the human being might then use the chatbot. We've learned that they use it a lot to develop verification work for them. That's where G2 is getting all of our traffic, because we just kind of confidential, confidentially, I'm on a podcast. Radix reports that 60% of the citations that come to G2 are coming from our Best of Guides. That's the Best of Software awards, all the listicles, all the grids, all of that stuff. Why? Because it prompt matches. Because the users are asking, give me the best three. And our Best of Guides win out over Blogger Best of Guides. Because our Best of guides are based on thousands and thousands, tens of thousands, hundreds of thousands of customer reviews and they're all verified. We require screenshots that you own the software. We have your LinkedIn profile. We know you're a real person. We really pulled away from Reddit because of that after the entity update, which we can come back and talk about, which changed everything. So for us, we found a way to be part of the verification layer work the models do, because the models now separate training research teams based on use cases that connect with their business model. So when you look at OpenAI Jordan 2027, 2028, it's going to be clear as a bell that their top two sources of revenue are ads and shopping. That's it. They're not going to try to win the API token business for developers. You're going to build a big shitty business if you do that with Low margins. Their multiple has to be like Meta meets Amazon for them to grow into their fundraising. Why is this important? Because if shopping is going to be one of the top two revenue streams, then they peel off teams that just focus on returns that give purchase recommendations. And the dominant methodology they use for tuning for that is to avoid regrettable purchases. Now for some of the listeners here, regrettable. I've heard about this before. Yes, you have. Facebook's turnaround came from an initiative Zuckerberg led. He got it. A really great set of insights from some experts that regrettable minutes on Facebook or any platform, Instagram, you name it, regrettable minutes are the number one signal of platform deterioration. So if you reduce regrettable minutes, you get more loyalty. Same occurs in the labs at places like OpenAI and Gemini. They want to avoid regrettable purchases because regrettable purchases will either create switching or reversion. They want this to be a one way door. Once you start using ChatGPT to buy stuff, you never stop. That's like really important to them. What does this mean? They've changed the criteria by which they recommend a product. Those are called commercial citations. It's a different ballgame than background citations when you're just asking it to give you the history of something that is really low risk. That is a different. It's an entirely set of different neural network activities. So anyway, long answer, but G2 is really dialing into that and how we can win the verification layer work.
- [18:51]
Jordan Cooney
I mean, I want to come back to this whole shopping and buying experience in ChatGPT or AI in general. You know, we're recording just about a week after, you know, ChatGPT's announcement of ads and starting to pilot ads. So that's very exciting. But I want to come back to something that you mentioned earlier, which is G2 and citations. That G2 is one of the most prominent visible cited resources when it comes to these B2B buying decisions.
- [19:20]
Tim Sanders
Almost 70% of B2B citations. Yeah, yeah.
- [19:23]
Jordan Cooney
And this is I think a remarkable component because clearly there's a trust factor with G2.
- [19:29]
Tim Sanders
Right?
- [19:29]
Jordan Cooney
And clearly that has been a signal that you guys have built and you shared this already through years of verification process trust with buyers. Here's my rub question, which is I think the major challenge that a lot of brands and businesses have in this AI discovery world, which is if you're giving that information to an AI model, if you're building these guides as you referenced and they're citing and sourcing these, what's the purpose of G2, why would a user ever really engage on G2 when I get all the information right in an LLM response?
- [20:09]
Tim Sanders
Well, you get the information. You just still need to verify the commercial citation. So let me give you this. So here's a way to think about it. It we're going to have what we had a million human visitors from large language models in 2025. That's a big number. Huge number. Okay, and it was x higher than 2024 and it'll be x higher in 2026. Why is this the case? Not all citations are the same when it comes to traffic and click through. They're not the same.
- [20:37]
Jordan Cooney
They're not.
- [20:37]
Tim Sanders
So common knowledge would say that a background citation, that's everything but the buy these 3 CRMs part of a return. Everything else is background citation. You don't have to verify it as a user because there's no risk. We've learned to trust it. We believe it. That's why the click through for a background citation is probably one quarter of 1%, maybe a third of a percent on perplexity, which is not a player as much as they used to be, but it's a quarter of 1%. You're not going to see any meaningful traffic. Early research and we're going to do real research on it this year. Early research on commercial citations like these are the 3crms you should buy. And the citation for the 3crms is G2 Profile. That's the citations we get. Okay. That click through rate is somewhere between 3 and 15% depending on the price tag, the risk factor and the nature of the user's distrust or trust in the model. So the answer is they get everything they need from the model except verification layer. They click on the citation to get verification layer. That's why software review sites are a close second to AI chat for everything from creating the shortlist to making the final decision. So the users trust would verify and that second, what's making us part of the game. I think that for freemium type software they may directly then make a purchase. What you also have to understand is important here is that many returns, I mean many like the majority of returns for a purchase recommendation don't link to the vendor unless the vendors in a shopping deal with OpenAI, they'll link to the proof point, the review site, the Gartner point, the blogger, because again, they're trying to avoid a regrettable purchase and they want to make sure that you are satisfied with your purchase. Lot of times what we're seeing is we get the traffic that the vendor used to get an organic search. But the traffic that comes to us from language models is much higher conviction. They're much further down the cognitive process. So then if they come to a software Vendor's profile on G2 out of that environment to read the reviews and then they click to go to the software vendor, the vendors are telling us much higher closing rate, much higher quality traffic. So anyway, that's the long and the short of it is that the user doesn't trust enough to then stop the session, go to the trouble of looking up the vendor side and contacting the salespeople. They still do some verification work and click through on that piece of the puzzle. And the other thing too I'm going to add is that G2 is beginning to add. In addition to reviews, we're going to be adding more and more benchmark data that actually measures things like hallucination rates and tool call reliability and other such technical metrics for agents. And because agents is a fast growing category, so we're seeing that that layer is going to be really important as well for the verification where models are going to love benchmarks and they already do when they recommend different AI solutions.
- [23:25]
Jordan Cooney
So I mean I love this frame of thinking, especially in how a user or a buyer is using this trust and verify process through a model. And a response here at Pre Visible, we did a study of 2 million LLM sessions, breaking them down by category. What ended up happening, how many events that triggered. And I think one of the most remarkable findings of that is the ratio of events to clicks. So when a user does go and click, they have a much higher degree of activity on the site, whether it's signing up for newsletter, not always buying. Right. I think we're always thinking about the final conclusion, but there's many sequences in between there. And I want to come back to that because this event and these events I think are the verification that the trust is there. Right? There's a reason why I'm going to follow that brand or I'm going to download the report or to your point on this benchmark piece, then I'm going to pull down that benchmark, I'm going to save it and I'm going to use that as a reference point as I continue my journey of making this buying decision. So I want to talk more about events and how G2 is a part of that event sequence in that citation that they were discovered in in an LLM.
- [24:44]
Time for a one minute break to hear from our sponsor, Pre Visible. So you're Looking for SEO help. Then you got a couple of options. You could start replying to spam from agencies that claim they can get you to rank number one on Google. You can pay an hourly rate for a consultant who will inevitably nickel and dime you with hourly charges. Or you can work with a cookie cutter agency to quickly launch a strategy less project with low success rate. None of those sound very good now do they? Well, that's where Pre Visible's integrated consulting model comes in. Pre Visible draws From a collective 40 years of SEO and digital marketing experience to unlock your organic growth opportunities. They build custom solutions that combine strategy, technical expertise, content and reporting to effectively operationalize SEO for your business. Pre Visible's four stage approach ensures that your SEO programs thrive by starting off with a strategy first approach. Then they support you in your efforts to create quality content, help you identify technical issues, and most importantly, they'll work with your cross functional teams to integrate your SEO strategies to make sure that your SEO budget actually drives results, not just your agency's bottom line. So join brands like Yelp, eBay, Canva, Atlassian Square, all who rely on the SEO consultants at Pre Visible. For more information, go to Previsible IO. That's previsible P R E V I S I B L E I.
- [26:13]
Tim Sanders
Yeah, that's fascinating. And, and, and so let's think about it this way. I'm sure if you've ever seen the Brian Johnston Don't Die documentary, you follow any of this stuff. You hear people talk about the difference between a biological age and a calendar age. A calendar age is like, you know, I'm whatever years old based on the calendar. Biological age is I'm whatever years old based on like my bio statistics. Right? So those aren't always the same. One of the ways to think about the buyer jerk learning is research indicates that when the buyer comes to you as a software seller, they're coming to you earlier from a calendar standpoint, earlier in the process. But from a cognitive standpoint they're much further along. So it used to be because of how long it took to like do the Google Blue link hell and build the shortlist and have the meetings and arrive at the three and then actually go contact the three. That might have been like a 13 day process, but now it's a three day process. So three days in they're talking to the vendor, but they're 90% through the bundle, meaning they've already got a short list, they've already got a rubric for how they're going to decide and they're just trying to do some tie breaking. That's what we see when you think about event. So that person is not. They're fresher, so they haven't started to experience fatigue. So they are going to respond more to novel offers that may not necessarily be part of verification works. What makes all the sense in the world that you're seeing them signing up for webinars and newsletters because they're fresher in the process and have more bandwidth to give it. They're actually still curious like, like nothing kills curiosity like time. So when you're further along in a calendar basis in a buying cycle, by the time you get to the vendor, you're like, damn it, just give me the, the contact information. At this point, I'm so tired of the subject. So I think that's a fascinating thing to think about is the difference between when buyer journey calendar age versus cognitive age. Because we do know that if you're using the model for synthesis, you're just much further along cognitively. And that doesn't mean you're set in your ways, but it does mean you're just much further along in developing an interest and an understanding deeply of that, of that purchase activity that you're in.
- [28:18]
Jordan Cooney
This is a really a critical piece of advice here, Tim. Largely because it's not very well documented how we think about users or buyers and how we target them. We bucket these individuals in audiences or Personas, right? But when we really think about where they're at in that phase of the decision making, this biological versus calendar perspective of where they're at in that buying journey, it can largely predicate how we think about our content strategies, what the messaging should be, how we build our web pages and our experiences and then right where we should be targeting them. Talk about these strategies and let's dive into like what are you seeing working? How are you understanding how brands are being successful in reaching consumers through AI discovery? What are some of the maybe tactics. You mentioned one of them earlier that G2 is leveraging to bring users into their journey to help them better be informed about B2B buying decisions.
- [29:26]
Tim Sanders
So when you think about what makes a model really effective, it's weight weights. I mean that's the secret, secret to chat GPT 5.2 is its weights. That's the secret to Gemini 3. Whenever you look at model performance, it's like how they weigh, you know, how, how inference is influenced by weights which the humans determine based on all the data. That's the secret sauce. Okay? That's the Secret sauce for marketers. How do you weigh traffic? What weights do you put in to your activities based on the source of traffic? So there's a sophistication we have to have now almost like a model model and how we assign weights based on where the traffic came from and our, our inference about how far along they are in the process. Because of what we're really going to understand is not all traffic is equal. That's why, you know, when I see similar web and other stuff that's trying to measure Chat GPT's influence based on traffic, I go, are you kidding me? No, that, that's not measuring its influence. Right. So I think, I think that we have to make sure that we don't have a monolithic view of traffic. I mean, let's go back G2. Our traffic went down last. Not money, citation work, like content, like the content that we produce in the category page, top level. It's like we all suffered traffic. We turned it around. I can tell you in a minute we're actually going to have more traffic now this year than we did last year. That's amazing. For us, that's like was a major thing to figure out. We figured out a few ways to do it. Most of that traffic recovery is not only going to come from Chat GPT and Gemini. It's going to be higher quality traffic for our buyers. But we figured that out and we can talk about that at a granular level. But weights are everything. Think. Let me give you like another like neurospicy piece of advice that'll really help the folks here. If you run an audit for AI crawlability on your site. Yeah. It's going to come back and say you're blocking this, you're slowing down that. And you've probably already done that. What you might be surprised is it's going to come back and say, hey, these gated reports are keeping the models from indepth our models from crawling and then referring back to your reports outside of deep research sessions. And this is increasingly true because 5.2 and Gemini 3. I'm afraid of ChatGPT and Gemini. They become kind of compute stingy. Now I know they want to avoid a regrettable purchase. Now they do index by giving more compute to an enterprise user than a consumer user. But if you're coming in on ChatGPT free, you get very little compute. And they ain't looking at PDFs on a vendor's website.
- [31:52]
Jordan Cooney
Website.
- [31:53]
Tim Sanders
No. Okay. But you say, no, we gated it, we made it a PDF for these reasons, we gated it for lead gen. We made it a PDF because the PDF, it just goes viral inside an organization by email. What you have to do now is start to use the concept of expected value now when you make fundamental marketing decisions. So let me, let me reach back. This is not a background. This is my bookshelf. Okay folks, books, I apologize. It's a thick book. It is a great book. If you haven't read this book and I make no affiliate, it's a joke for those Reverend Ignore affiliate affiliate link in the in the show notes called on the Edge By Nate Silver natesilver538.com Superforecasting Also the guy that went to the final table at World Series of Poker earlier in his career by using EV on the edge. It's about having an edge in every decision by using expected value calculations to take the emotions or the obsolescence out of a decision. So I've been working with marketers that say, okay, we need to run the EV on a gate and look at the expected value of the leads versus the expected value of AI visibility and make a better decision. So what's going on in a lot of organizations is maybe they've got a lot of JavaScript, JSON, JavaScript required rendering pages terrible for AI crawlability, but you're going to have to spend calories to fix that and refactor those pages. You're going to run EV on whether it's worth it. You might have I mentioned before PDFs that need to be republished in HTML or even better markdown. We've been doing that a lot at G2 across pages and seeing amazing AI visibility, chat, GPT traffic come from that. All of that takes calories and expenses to do. You don't just do it because AEO is hot, you do it because the EV runs positive. To do that, the expected value is actually greater than the cost. So I think more than ever we need to become masters at EV. But here's the good news folks. ChatGPT and Gemini are great at doing low hallucination EV work if you give it the right context.
- [33:54]
Jordan Cooney
How frequently in this world where we're making these EV decisions right, and we're trying to become incredibly targeted to certain buyers and users at a phase in their journey. How are you thinking about about the outcome metrics, the actual metrics that happen post the evaluation period? So let's say you guys run this campaign of building all these benchmark studies. You're going to see that there's a value that there's an expected value that you're calculating, that there's something you want. But then there's this post regression of like, hey, did this actually do what I wanted it to do? How frequently do you think marketers need to be doing this, this and at what variable are you making these kinds of changes in what is largely a decision making process that maybe only happens once a year or maybe once every few years when it comes to B2B buying.
- [34:51]
Tim Sanders
That's a great question. I'm going to go off on a limb here and talk a little bit about Toyota Lean and the great recalibration of how often they recalibrated machines inside manufacturing. So this goes back to like gosh, the 1970s, believe it or not, at Toyota, a high number of brakes didn't match. And so they had to do a lot of rework and it was a source of great quality. And what they found out was by using the five Y's, which is a great technique to get to the root cause of a problem, they kind of got to the bottom of it. And that was there's a recalibration you do in the factory, in the lathes, and you recalibrate it back to specification. And what happens is like if the recalibration period's too long, then it starts to kind of drift a little bit and all of a sudden the brakes don't match. But the question was, do we recalibrate every day, every week, every month, every quarter? And that's where the statistical analysis was. Well, it depends on how fast the lays really move. So what they had to kind of get to is what's the rate of movement of these machines to vary from specification. And once they knew the rate of movement, they're like, oh great, we only have to do it every two weeks. They would have dramatically overspent. If they did it daily, they would have undershot the Six Sigma quality goals. It wasn't called Six Sigma at the time if they did it every two or three months. So the big takeaway because used to work for Dr. Edward Deming. So I was like part of this whole Six Sigma movement. The takeaway from all that is that what you have to understand is like how fast the variables are moving. Because once you get an understanding of the variable moving speed, then you understand what the recalibration requirement is. Same works now for marketing. How quickly is your market adjusting? So let's go back to the benchmarks question. The reason G2 started the benchmarks is because our research starting in 2024 led us to believe that as AI moved from experiment sustainable think of it like WI fi and electricity in an organization, it would transfer from being a business buyer like SaaS to becoming a central IT buyer like pre SaaS. And that's what we saw coming. And so in 2025 the number jumped where it looked like about 50% of all AI purchases were originating out of central IT who is now writing the spec. And then our latest survey in August showed that like 70% of all agents, their economic buyer is going to be it. The business buyer is going to be their internal customer. So I went out and did over 40 interviews of IT leaders over the very short period of time to gather insights. And so we saw, oh my God. Within a few years the machines are moving to the point where NPS signals like reviews may not be sufficient for trust and verification layer work. Because all of our interviews revealed that central of IT cared less about NPS signals like ease of use or stated roi, which they think is not an answerable question. Or it's customer satisfaction that doesn't matter. Here's what matter. The benchmark for hallucination, the benchmark for tool call reliability, the benchmark for containment rate. That stuff matters to us. And we knew this because they spent all their time on like seal leaderboards or Elle Marina. So we begin to see that movement. Well the good news is that this is not going to go back and forth like the business isn't going to take back buying software next year. So it looks like the machines are moving rather slowly. So for us, I would suspect the cadence of remeasuring that might be annual or at the most every six months. Now that's not the case for every business. Some businesses are seeing really fast moving situations. Like I think a great example, CRM. CRM is a super dynamic space right, right now. And so as a result the machines may be moving every quarter. So anyway, the long answer here is you have to really have your finger on the pulse of how quickly biomores and criteria are shifting and calibrated accordingly.
- [38:39]
Jordan Cooney
You brought up something in there that really, I mean a lot of that resonated by the way. But one of those components that I want to glean out for our listeners is, is the value of a review. And there's, there's a lot shifting underneath us in this concept. And I think that over the past year, looking at 2025, the value of a review dramatically rose right as LLA.
- [39:04]
Tim Sanders
Especially a recent review. We can talk about that in a minute. But review recency in particular has risen a lot because it reflects. Because the models value recent reviews because they reflect the latest capabilities.
- [39:17]
Jordan Cooney
No question. And I fully agree. Recency quality signals like what you were talking about, like verified or trusted types of reviews.
- [39:27]
Tim Sanders
I'll share some inside baseball token link matters now like a longer review that shares more detail that might share what the trepidations were going in, how they were. That feels more like a story with an arc is going to sync better with the weights of both OpenAI and Google DeepMind better too. Yeah, so that's interesting. And we've been moving to a. We bought a company last year called Unsurvey and they've had that interviewing process where the AI interviews a person. It can get a much richer conversation much more quickly. And then we turned on voice where you can just say the thing and it's dramatically. I can't tell you to how how much, but it's a lot. Increased the tokens inside the reviews and created reviews that feel more like storytelling than blurbs. Like the back of a book. Like five, you know, buy this book. It's great. We wanted to get away from that because the models don't value that as much.
- [40:24]
Jordan Cooney
And I love this because this is, I think one of the big crossroads that we're facing in 2026, which is there is review, review frequency and review value. But we're also quickly moving and shifting because these models are ingesting far more data and information that also contains valuable insights. You know, things like podcasts, what we're doing right now, video and video information about products and tutorials. And so this whole multimedia component is being layered on top of these review stacks. And I want to get your perspective in terms of where the future is going. There, there.
- [41:03]
Tim Sanders
Okay. So if you look at the chart for B2B, just look at B2B software queries and you see G2 leads by a mile. But you know, right behind that is YouTube. You wow. YouTube is a source. Why? Yes, first of all, many YouTube videos have a transcript. It's very important for OpenAI for chat GPT that you have a transcript. They're not going to watch your video. Gemini might. Chat GPT is not going to Chat GPT is going to read the transcript. But the reason video is so compelling, Jordan, is that it becomes a forcing function for marketers to create answer shaped natural language content as opposed to message shape, corporate shape content. Okay. So when we're creating landing pages, we have a messaging approach, problem solutions, feature, benefits, proof. But when we're writing an faq, we have compelling question answer. Think about our conversation today. Think about the average video you might make with a customer. It's very much Q and A. So it creates that very answer shaped content the models prefer in verification work. They're actually trained to understand that better. It also fits better within their allergy to corporate vendor messaging that might create a regrettable purchase. So I love YouTube video because you can't make corporate messaging out of one of these conversations. We're just talking to each other. You can't push me back to say the thing like your marketing team wants to always say the thing that the models are like. They're just trying to say the thing. So YouTube and podcast are a forcing function for you to do what I consider one of the most three important things you do to win in AI search world and that is to refactor content to be answer shaped instead of message shaped and to incorporate more natural language in what the models see when they come to you.
- [42:52]
Jordan Cooney
Love this. I want to bring up one but.
- [42:54]
Tim Sanders
You have to produce the transcript. Make sure the transcript is visible to the bot or you've missed it completely.
- [43:00]
Jordan Cooney
I mean accessibility as a whole is one of the major trends and challenges that I think all of these models are going to have. And, and in a landscape now where it's not a single monopoly, right. It's not just Google and Gemini but it's going to be multi pronged to some extent. Right. It's going to be ChatGPT and Google in some instances. I do think there's probably buyer journey instances where Claude is going to be very heavily influenced as opposed to that.
- [43:26]
Tim Sanders
Yeah, yeah.
- [43:26]
Jordan Cooney
And so they're 5% copilot.
- [43:28]
Tim Sanders
Don't sleep on copilot. Enterprise buyers, they get guard railed. All they have is co pilot chat and they're enterprise buyers. And I wouldn't sleep on it now. And, and, and, and, and I, I think of co pilot like from a fashion standpoint the Canadian version of chat GPT Canada always gets it just like you know, 18 months later. So, so even though copilot is not as, as rich in inference as getting 5.2 GPT, it's close. So a lot of the things you figure out for chat GPT will help you in co tell AEO and SEO marketers all the time like have a, if you're enterprise and you sell to enterprise people then you need to treat co pilot like, like an analyst. It may not be Gartner, it may not be forester. Okay. It's IDC treat it like an analyst, respect it, study it, understand how it lives and breathes.
- [44:16]
Jordan Cooney
Yep, a hundred percent. I have to, I have to close out our, our, our main part of our episode with this major question because I, I, I fundamentally think it's something that every traditional SEO, the new aspiring Geo AEO marketers are all thinking about and trying to understand. And that is that when you think about these tactics and you think about whether it's how you influence reviews, how you increase your visibility, say in video or video testimonials, there's this habit as humans to be like water and find the path of least resistance. And we try to go into the game of thinking of in the frame of mind of how do we manipulate these communities, how do we influence them and change their perspective? How do we get our brand positioned in a way that oftentimes is not genuine or is not factual or is not real to the experience of the product or software? And we're seeing that now with the influence on Reddit and how much people are trying to gamify or change Reddit threads and communities. I'm sure G2 has strict policies on this and verification, as you mentioned earlier, of not just the reviews but the content. As we think about this year ahead, how do we think about really creating genuine campaigns and efforts, efforts that reflect what human buyers want.
- [45:49]
Tim Sanders
Well, I'm going to make another book recommendation here too. So Seth Godin wrote this great book a couple of years ago called this is Marketing and he said best marketing message you can ever have is the following People like you do this. So I think that what hasn't changed your Jordan is the fundamentals still matter and if you market from a place of empathy that what you're trying to do is help the right buyer find the right product and never regret the purchase and you make that your North Star, you're aligning with something that will probably be as endearing, especially in the purchase recommendation use case as Larry Page's PageRank system, which to this day is still influential. Right. So this kind of gets to I wanted to make sure and point this out too is like like SEO has never been more important because real time research still relies on ranking to some extent outside of deep research. The fundamentals of SEO are transferring to success in AEO to a large degree and certainly enough for us to continue to value it. We're treating AEO as a new swim lane, not a replacement. So I want to make that point. And so this idea of originating from a place of empathy has never been More important, here's what's happening, though. You won't be banned from all the large language models. Like you can be banned from Google. That's the good news. The bad news is the latency between finding a hole in the system and it getting closed is shortening a lot. Like with Google, I saw things that occurred where these content click farms had like two years unfettered success. And then Google makes a change and they die. It ain't gonna be no two years. It's one model to the next. Because regrettable purchases are super easy to verify. For those of you listening, go Google the concept of Verifier's Law. Verifier's Law says anything that's easy to verify will be conquered by AI. Love it. So regrettable purchases, easy to verify. Just tell the truth. If you work at a company that I offers a crappy service, find a new job.
- [48:08]
Jordan Cooney
All right, Love this. We're going to transition now to our lightning round. I'm going to ask you five questions from the topics that we discussed today. Quick response in about 30 seconds to a minute on each. All right, tough.
- [48:20]
Tim Sanders
Bring me. Here we go.
- [48:21]
Jordan Cooney
All right, here we go. What's the one buyer insight from G2's data set that most surprises marketing teams?
- [48:29]
Tim Sanders
That the executives no longer drive the purchase of artificial intelligence because they've moved on to transformation and adoption. So focus on the CEO may be a miss. You may want to focus on the grunts in central IT. Yeah, man.
- [48:45]
Jordan Cooney
Knowing that audience, if you could change one common B2B, go to market assumption about the whole new AI discovery world, what would that be?
- [48:56]
Tim Sanders
That keywords are the same thing as prompts.
- [49:00]
Jordan Cooney
Tell me more. Tell me more. That is so powerful.
- [49:02]
Tim Sanders
It's a huge prompt problem. Right? So the use of keywords will no longer be a real reflection of how people in an answer engine environment prompt AI. And if you go to providers like Scrunch who actually have panel data, people's prompts, you'll see such misalignment between, like how a person asks for a solution. Like, let me give you an example. Recently I adopted Willow Voice. It's what the cool kids call voice keyboards. So I just click function, I say the thing. It's perfectly structured. It goes out. I don't type. I'm three times more productive. It's bigger for me than ChatGPT. But when we talk to one of their big competitors, the competitor's like, nope. We call it AI voice dictation. And they just stick with it. And then they go in and they look at dictation, they buy all the keywords and everything, and then we go get the prompt data and everybody in the icp, which would be all the IT people, that would lead to all the adoption across the enterprise. They strictly call it voice keyboard. So there's a tremendous misalignment, by the way, all the voice it, all the IT cool kids, they never use Google. So you're not going to see it in AdWords, in the keyword research, you're going to miss, miss it. So the problem is, like, all the early adopters have moved to Chat GPT and they find their own way of talking about a thing in the prompt world that may not show up in the key world. Ward, if you rely too much on keywords and aeo, you're going to miss the mark. And worst of all, you're going to create false positives where because you're using the way you talk about it and the keywords you're focused on, you're going to think you're doing better than you are on your share because of course you're going to win that which you engineer for. So that's one thing. And I could share change.
- [50:36]
Jordan Cooney
Love it. Love it. Does AI visibility ever hurt a brand's conversion funnel? And if so, how?
- [50:44]
Tim Sanders
I don't think so. I don't think so. I don't think that AI visibility is going to make you less credible. I don't think coming back as One of the three recommended CRMs is going to hurt anything now if someone else in that company is already in flight, it's just another verification signal for the humans internally. So I don't think there's. And I've heard about this before, especially from the vaunted guards of the old way. No, the answer is the fact that ChatGPT recommends something is not going to cause infosecurity to say, well, it must suck. Now, that being said, the way we talk about using AI can be bad. In other words, if a person says, I put together all my purchase recommendations with ChatGPT, there is research that might indicate people around the table are going to roll their eyes and say, lazy. It's slop. So there is the risk that the way we represent the AI research can cause it to be perceived as slop and discounted, especially by those that are in my category of boomer age. But that's more of a skills issue about how you talk about it. I'm rather transparent in how I use Chat GPT. I'm responsible for my work I check for hallucinations, but I never leave. But I just plug it into ChatGPT because I know what the old folks think about that. Yeah.
- [51:54]
Jordan Cooney
All right, we got two more here. What's the most underrated signal that enterprise brands should optimize for in AI responses?
- [52:04]
Tim Sanders
I think it's not so much of a signal, but it's a signal to the bots. I think that markdown language key takeaways from a page that are placed at the very top of the page are going to create the most crawling and reference and training inclusion of your web pages than anything. So when you think about a really. And I'm talking about sales pages, I'm talking about product pages. So a small markdown file with the key four takeaways will dramatically improve the chances that your content is considered during test time.
- [52:42]
Jordan Cooney
One that I think we didn't really talk about much, but is pricing pages. It's so complicated for business buyers, and you're like, I don't know what this is and how do I. How do I measure the cost of this?
- [52:54]
Tim Sanders
Sometimes, sometimes, sometimes the opaqueness of pricing gives the leverage to the. To. To the seller. But you need to run EV because you have to now account for the fact that it creates confusion for the model. So this is another classic example of where expected value might, like, liberate you from, like. No, no, no, no. Read the book. Never. What is it called? Never split the Difference by Chris Boss. Like, we got to use negotiation tactics. Maybe not.
- [53:16]
Jordan Cooney
Yeah, love that. All right, last one here. If you could add one new media form to be more visible in AI platforms, what would it be?
- [53:25]
Tim Sanders
It would be video with transcripts for videos. Okay.
- [53:29]
Jordan Cooney
And tell me why. What makes transcripts for video so valuable?
- [53:32]
Tim Sanders
They're answer shaped or natural language. They're human, they are perceived as real, and they match the language training for most models that goes back to the beginning of a model. So it's deep set, awesome. It helps you get out of your own way as a corporate market marketer.
- [53:49]
Jordan Cooney
Yeah, absolutely right.
- [53:51]
Tim Sanders
Like, who corrects a transcript layer to make it feel more consistent with our corporate brand bible? No one. That's why the models love it.
- [53:59]
Jordan Cooney
Yeah, that's great. Okay. And that wraps up this episode of the Voices of Search podcast. Huge thank you to Tim Sanders, Chief innovation officer at G2, for joining us. If you'd like to contact Tim, you can find a link to his LinkedIn profile in our show notes, or you can find it on voicesofsearch.com you can also visit his personal website Tim Sanders.com if you haven't subscribed yet or would like a daily stream of SEO and content marketing knowledge in your podcast feed, hit the subscribe button in your podcast app or on YouTube and we'll be in your feed next week. Okay, that's all for today, but until next time, remember, the answers are always in the data.
- [54:41]
Tim Sanders
Sam.